[
https://issues.apache.org/jira/browse/FLINK-6293?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15988390#comment-15988390
]
Till Rohrmann commented on FLINK-6293:
--------------------------------------
I think the problem is that in the "handle trigger savepoint response for
non-existing job" test, we retrieve the leader gateway but do not wait until
the JobManager has gained leadership. This is possible when using the
standalone leader retrieval service. As a consequence, we can end up sending
the {{TriggerSavepoint}} savepoint message too early (before the JobManager has
gained leadership and, thus, dropping the {{TriggerSavepoint}} message).
> Flakey JobManagerITCase
> -----------------------
>
> Key: FLINK-6293
> URL: https://issues.apache.org/jira/browse/FLINK-6293
> Project: Flink
> Issue Type: Bug
> Components: JobManager, Tests
> Affects Versions: 1.3.0
> Reporter: Nico Kruber
> Assignee: Till Rohrmann
> Priority: Critical
> Labels: test-stability
>
> Quite seldomly, {{JobManagerITCase}} seems to hang, e.g. see
> https://api.travis-ci.org/jobs/220888193/log.txt?deansi=true
> The maven watchdog kills the build due to not output being produced within
> 300s and {{JobManagerITCase}} seems to hang in line 772, i.e.
> {code:title=JobManagerITCase lines
> 770-772|language=java|linenumbers=true|firstline=770}
> // Trigger savepoint for non-existing job
> jobManager.tell(TriggerSavepoint(jobId, Option.apply("any")), testActor)
> val response = expectMsgType[TriggerSavepointFailure](deadline.timeLeft)
> {code}
> Although the (downloaded) logs do not quite allow a precise mapping to this
> test case, it looks as if the following block may be related:
> {code}
> 09:34:47,684 INFO org.apache.flink.runtime.minicluster.FlinkMiniCluster
> - Akka ask timeout set to 100s
> 09:34:47,777 INFO org.apache.flink.runtime.minicluster.FlinkMiniCluster
> - Disabled queryable state server
> 09:34:47,777 INFO org.apache.flink.runtime.minicluster.FlinkMiniCluster
> - Starting FlinkMiniCluster.
> 09:34:47,809 INFO akka.event.slf4j.Slf4jLogger
> - Slf4jLogger started
> 09:34:47,837 INFO org.apache.flink.runtime.blob.BlobServer
> - Created BLOB server storage directory
> /tmp/blobStore-eab23d04-ea18-4dc5-b1df-fcf9fc295062
> 09:34:47,838 WARN org.apache.flink.runtime.net.SSLUtils
> - Not a SSL socket, will skip setting tls version and cipher suites.
> 09:34:47,839 INFO org.apache.flink.runtime.blob.BlobServer
> - Started BLOB server at 0.0.0.0:36745 - max concurrent requests: 50 - max
> backlog: 1000
> 09:34:47,840 INFO org.apache.flink.runtime.metrics.MetricRegistry
> - No metrics reporter configured, no metrics will be exposed/reported.
> 09:34:47,850 INFO
> org.apache.flink.runtime.testingUtils.TestingMemoryArchivist - Started
> memory archivist akka://flink/user/archive_1
> 09:34:47,860 INFO org.apache.flink.runtime.testutils.TestingResourceManager
> - Trying to associate with JobManager leader akka://flink/user/jobmanager_1
> 09:34:47,861 INFO org.apache.flink.runtime.testingUtils.TestingJobManager
> - Starting JobManager at akka://flink/user/jobmanager_1.
> 09:34:47,862 WARN org.apache.flink.runtime.testingUtils.TestingJobManager
> - Discard message
> LeaderSessionMessage(00000000-0000-0000-0000-000000000000,TriggerSavepoint(6e813070338a23b0ff571646bca56521,Some(any)))
> because there is currently no valid leader id known.
> 09:34:47,862 INFO org.apache.flink.runtime.testingUtils.TestingJobManager
> - JobManager akka://flink/user/jobmanager_1 was granted leadership with
> leader session ID Some(00000000-0000-0000-0000-000000000000).
> 09:34:47,867 INFO org.apache.flink.runtime.testutils.TestingResourceManager
> - Resource Manager associating with leading JobManager
> Actor[akka://flink/user/jobmanager_1#-652927556] - leader session
> 00000000-0000-0000-0000-000000000000
> {code}
> If so, then this may be related to FLINK-6287 and may possibly even be a
> duplicate.
> What is strange though is that the timeout for the expected message to arrive
> is no more than 2m and thus the test should properly fail within 300s.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
